how many people run with JS disabled?

[Lee]

RobG said:

RobG wrote:
[...]

And there you have it. Were they also statisticians and
suitably motivated, they would have devised appropriate
measurements and actually *calculated* the error in the
statistics.

I know statistics. Margin of error calculations require that the
sample population be a random sampling of the actual population.
In such a case, the error will be due to the sample size being too
small.

In this case, a large portion of the error is due to systematic
sampling error. No amount of number crunching can correct a
poorly designed sampling method.

 

[Mick White]

Lee wrote:

I know statistics. Margin of error calculations require that the
sample population be a random sampling of the actual population.
In such a case, the error will be due to the sample size being too
small.

In this case, a large portion of the error is due to systematic
sampling error. No amount of number crunching can correct a
poorly designed sampling method.

Well, let's design a better model, meanwhile we could use a little
common sense. If js is vital, let the user know it, if not, accommodate
the Luddite.

Statistics or no, I can confidentially assert at least 95% of the users
of my sites have js enabled.

That statistic is important to *me*, extrapolation I leave to the
statisticians.

Mick

 

[Randy Webb]

I can see how things like caching and IP address ambiguity leads to
wrong results (if absolute counts is what you're after), but I don't see
why percentages derived from http statistics (e.g. browser type,
javascript availability etc) should be so badly off, especially if a
large number of samples is looked at.

Scripting enabled/disabled is a little easier to track than browser type
is simply because of spoofing. The userAgent String that Mozilla gives
me with the prefs bar set to spoof IE6 is exactly the same as the
userAgent string given to me by IE6. So a server has no way of knowing
whether I was using IE or Mozilla, and that alone makes the statistics
based on those logs worthless and inaccurate.

Another problem other than caching and proxies has to do with the way
browsers make requests instead of how HTTP works. I have a test page
that shows the following requests:

IE6: 128
Opera 7: 1
Mozilla: 1

What percentage of the requests were made by each browser?

IE6: 1/3
O7: 1/3
Mozilla: 1/3

I know those numbers because I made the requests myself.

Bonus question: How many images are on the page I requested?

 

[Randy Webb]

RobG wrote:
[...]

And there you have it. Were they also statisticians and
suitably motivated, they would have devised appropriate
measurements and actually *calculated* the error in the
statistics.

But the reason they don't calculate that margin of error is the same
reason that the statistics weren't any good to start with. It's
impossible to determine, even with a margin of error.

I beg to differ. I think it is possible to estimate the error,
though I agree that collecting data from a single server is
unlikely to produce reliable results. But...

Read my other reply in this thread and see if it makes sense, and, if
you can answer the bonus question. There is more to it than a simple
margin of error.

That is your opinion, which is only half the argument. The
other half is whether applied mathematics can create a model of
the system and accurately predict outcomes based on data
collected.

The only way it could even come close to that is to know all, and I mean
*all* of the variables and thats impossible to know. If I have my cache
set to never check updates, and the next user has it set to always check
(or empty at browser closing), and the next has it set to....... And it
can go on and on. There is absolutely no way to even come close to
creating an "accurate" model of the Internet.

I do not doubt your knowledge of Internet systems, nor your
ability to apply that to problems within your realm if
expertise, but I find your lack of faith in statistical
modeling disturbing...

Statistic Modeling has my faith, applying it to the Internet doesn't.

<that needed a Darth Vader voice ;-) >

...so I'll bet you aren't a statistician.

Can't say that I am, but I know what they are, I use them daily, and I
know the flaws in the statistics I use.

No, it means you can't conceive a model that allows for them
(the issues).

And that is precisely why browser/internet statistics are worthless. You
can't come up with a margin of error without a model.

Measurements made and analyzed without regard for errors
inherent in the system will be useless, but the fact that you
claim intimate knowledge of those very errors means it is highly
likely that an accurate measurement system can be devised.

No, see above.

All that is required is a properly configured web page that
gets perhaps a few thousand hits per day from a suitably
representative sample of the web surfer population.

When I am at work sitting at my desk and request a web page from a
server, it does not go straight to the server. The proxy server that we
use is where the request is made to. From there the proxy requests it,
scans it and decides whether to let me have it or not. The only stat you
will get on the server is the ones from the proxy server. So, if I open
it, how will you determine what browser/UA I used?

 

[J. J. Cale]

Another problem other than caching and proxies has to do with the way
browsers make requests instead of how HTTP works. I have a test page
that shows the following requests:

IE6: 128
Opera 7: 1
Mozilla: 1

What percentage of the requests were made by each browser?

IE6: 1/3
O7: 1/3
Mozilla: 1/3

I know those numbers because I made the requests myself.

Bonus question: How many images are on the page I requested?

one is more than enough to generate anywhere from 7 to 15 hits since IE
deliberately causes extra requests to manipulate the statistics. I keep no
useable archives but if I recall we went round this once and I provided a
link to some support for this statement. It was in response to an image
loading question and research turned up the M$ bluff.
"lies.... damn lies... and [browser]statistics"
Jimbo
Jimbo